Re: [RFC PATCH v2 2/3] mm: handle large folio when large folio in VM_LOCKED VMA range

[Hugh Dickins]

On Fri, 14 Jul 2023, Yin, Fengwei wrote:
> On 7/14/2023 10:21 AM, Hugh Dickins wrote:
> > On Wed, 12 Jul 2023, Yin Fengwei wrote:
> >> On 7/12/23 14:23, Yu Zhao wrote:
> >>> On Wed, Jul 12, 2023 at 12:02 AM Yin Fengwei <fengwei.yin@xxxxxxxxx> wrote:
> >>>> --- a/
> >>>> +++ b/mm/internal.h
> >>>> @@ -643,7 +643,8 @@ static inline void mlock_vma_folio(struct folio *folio,
> >>>>          *    still be set while VM_SPECIAL bits are added: so ignore it then.
> >>>>          */
> >>>>         if (unlikely((vma->vm_flags & (VM_LOCKED|VM_SPECIAL)) == VM_LOCKED) &&
> >>>> -           (compound || !folio_test_large(folio)))
> >>>> +           (compound || !folio_test_large(folio) ||
> >>>> +           folio_in_range(folio, vma, vma->vm_start, vma->vm_end)))
> >>>>                 mlock_folio(folio);
> >>>>  }
> >>>
> >>> This can be simplified:
> >>> 1. remove the compound parameter
> >> Yes. There is not difference here for pmd mapping of THPs and pte mappings of THPs
> >> if the only condition need check is whether the folio is within VMA range or not.
> >>
> >> But let me add Huge for confirmation.
> > 
> > I'm not sure what it is that you need me to confirm: if the folio fits
> > within the vma, then the folio fits within the vma, pmd-mapped or not.
> Sorry. My bad. I should speak it out for what I want your confirmation:
>   Whether we can remove the compound and use whether folio is within
>   VMA instead.
> 
> I suppose you answer is Yes.

Yes (if it all works out going that way).

> 
> > 
> > (And I agree with Yu that it's better to drop the folio_test_large()
> > check too.)
> My argument was folio_test_large() can filter the normal 4K page out so
> it doesn't need to call folio_in_range() which looks to me a little bit
> heavy for normal 4K page. And the deal was move folio_test_large()
> to folio_in_range() like function so simplify the code in caller side.

I realized that, but agree with Yu.

It looked a little over-engineered to me, but I didn't spend long enough
looking to understand why there's folio_within_vma() distinct from
folio_in_range(), when everyone(?) calls folio_in_range() with the same
arguments vma->vm_start, vma->vm_end.

> 
> > 
> > This idea, of counting the folio as mlocked according to whether the
> > whole folio fits within the vma, does seem a good idea to me: worth
> > pursuing.  But whether the implementation adds up and works out, I
> > have not checked.  It was always difficult to arrive at a satisfactory
> > compromise in mlocking compound pages: I hope this way does work out.
> This is the purpose of this patch. :). Thanks.
> 
> 
> Regards
> Yin, Fengwei
> 
> > 
> > Hugh